Nolb Manual

User Guide
NOLB : Non-Linear Normal Mode Analysis, version 1.1, December 2018
Sergei Grudinin
Inria/CNRS Grenoble, France
email : sergei.grudinin@inria.fr
c 2018 Sergei Grudinin

Copyright
http://team.inria.fr/nano-d/software/nolb-normal-modes/
Contents
1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Introduction 3
1.2 NMA theory 3
1.3 RTB projection method 4
1.4 The NOLB method 5
1.5 Potential function 6
2 Program options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 Usage 7
2.2 Main options 10
2.3 Experimental options 13
2.4 Currently suppressed options 13
3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1 Analysis of basic molecular motions 15
3.2 Comparison with linear NMA 17
3.3 Another example with a bigger system that contains heteroatoms 17
3.4 Finding the best transition between two protein conformations 18
3.5 Analysis of an MD trajectory or NMR ensemble 19
3.6 Clustering of MD trajectory frames 21
3.7 Analysis of protein docking poses 22
3.8 Structural ensembles 22
3.9 Structural transitions 24
3.10 Nonlinear structural transitions 27
3.11 Analysis of modes 28
4 Related methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.1 GUI 30
4.2 Morphing paths 30
4.3 SAXS-guided structure optimization 31
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1. Theory
1.1 Introduction
Normal mode analysis (NMA) is an old and well established technique [1] that has recently found
many new applications in the field of structural biology and structural bioinformatics [2]. NMA
uses a quadratic approximation of the potential energy, and thus it produces linear deformations
of the initial structure, which are accurate only for small-amplitude motions. Larger amplitudes
can destroy, for example, the secondary structure and break interatomic bonds, when NMA is
applied to a protein. An obvious circumvention for this problem will be to take smaller amplitude
steps and iteratively recompute and diagonalize the Hessian matrix from the updated positions.
This procedure would indeed produce a more realistic deformation of the initial structure thanks
to the non-linearity of the obtained deformation. However, such an approach requires multiple
diagonalization steps, which may be computationally expensive for some of the applications. Thus,
multiple attempts were made to introduce non-linear deformations without the need of multiple
diagonalizations. Here we present a new scheme for nonlinear normal mode analysis in the
Cartesian coordinates that extrapolates a motion computed from instantaneous linear and angular
velocities to large amplitudes. The scheme can be considered as a natural evolution of the widely
used rotations-translations of blocks (RTB) method [3, 4].
1.2 NMA theory

Let us consider a molecular system with N atoms at an equilibrium position q0 ∈ R3N . Let
V : R3N 7→ R be the potential energy of the molecular system. Let us also introduce q ∈ R3N , a
small time-dependent molecular displacement of our system around q0 . The potential energy V in
the vicinity of q0 thus can be given by its quadratic approximation, which allows to analytically
solve the Newton’s equation of motion,
M(q̈ + q¨0 ) + ∇V (q0 + q) ≈ M q̈ + Kq = 0, (1.1)
where M is the diagonal mass matrix, and K is the Hessian matrix (or stiffness matrix) of the
potential energy V evaluated at the equilibrium position q0 . We then compute the square matrix
of eigenvectors L and the diagonal matrix of eigenvalues Λ of the mass-weighted stiffness matrix
Kw = M −1/2 KM −1/2 ,
Kw = LΛLT . (1.2)
Let η ∈ R3N be the projection of q into the eigenspace of Kw and let us call (λi )i=0...3N the diagonal
values in Λ. Then, left multiplication of eq. 1.1 by LT M 1/2 gives the following system of uncoupled
4 Chapter 1. Theory
equations,
η = LT M 1/2 q
, (1.3)
η̈i + λi ηi = 0 i = 1 . . . 3N
which can be solved by the classical ordinary differential equation (ODE) theory.
1.3 RTB projection method
A B C ~v = ~v⊥ + ~v|| !
~
!
~
~v = ~v⊥ + ~v||
Figure 1.1: (A) All-atom representation of a protein. (B) RTB representation of the same protein.
(C) Each rigid block has six degrees of freedom.
The main idea of the RTB projection method is to approximate the system as n rigid blocks.
Figure 1.1.A-B schematically shows the all-atom and the RTB representations of the same system.
The transition from the RTB coordinate system with 6n DOFs to the all-atom coordinate system
with 3N DOFs is performed by an orthogonal projection matrix P ∈ R3N×6n . The conservation laws
of the linear momentum and the angular momentum of a rigid block consisting of Nb atoms written
in mass-weighted coordinates are
Nb
√
Mb q̃˙ =
p
∑ mk q̇k for a translation
k=1
, (1.4)
Nb
√
I1/2 q̃˙ = ∑ mk (qk × q̇k ) for a rotation
k=1
where Mb is the total mass of the rigid block, I is the rigid block’s inertia tensor, q̃ is the blocks’s
displacement, mk is the mass of the kth atom of the block, and qk is the displacement of the kth
atom of the block. The elements constituting PT , the matrix projecting an all-atom motion q into a
motion of rigid block q̃ are then obtained by differentiating (1.4) with respect to q̇k [5]. This leads
to translation Pt and rotation Pr matrices of size 3Nb × 3 each, computed for each of the rigid blocks
and written through their k 3 × 3 square components,
r
mk
Ptk = E3 for a translation
Mb , (1.5)
√ −1/2
Prk = − mk (I) [rk − rCOM
]× for a rotation
where k is one of Nb atom indices, rk is the position of the corresponding atom in the block, and
rCOM the position of the block’s center of mass (COM). The rigid block’s displacement (δ , θ )
1.4 The NOLB method 5
6-vector is then obtained by summing up the displacements in the RTB coordinate frame,
Nb
δ= ∑ Pt Tk qk for a translation
k=1
. (1.6)
Nb
θ= ∑ Pr Tk qk for a rotation
k=1
Having written these equations, we can write the projection matrix P as a diagonal block matrix,
 1
Pt Pr 1

P = ... ... ... . (1.7)

 
Pt Pr
n n
The normal modes are then computed by the diagonalization of the RTB-projected mass-
weighted stiffness matrix,
PT Kw P = L̃Λ̃L̃T , (1.8)
where L̃ is the matrix composed of the RTB normal modes with the corresponding diagonal
eigenvalue matrix Λ̃. This equation can be rewritten as
Kw = (PL̃)Λ̃(PL̃)T . (1.9)
The all-atom normal modes (in mass-weighted coordinates) are then obtained as a projection of
Lw
the RTB normal modes L̃ according to
Lw = PL̃. (1.10)
1.4 The NOLB method

Molecular vibrations in a multi-dimensional harmonic oscillator are all uncoupled and can be found
by solving (1.3). Diagonalization of the RTB-projected mass-weighted stiffness matrix gives a set
of eigenvectors that are composed of instantaneous linear velocities and instantaneous angular
velocities of individual rigid blocks {~vw , ω
~ w }, see Fig. 1.1.C. For a rigid block with mass Mb and
inertia tensor I, we first compute these in the non-mass weighted coordinates as follows,
−1/2
~v = Mb ~vw
. (1.11)
~ =I −1/2 ~
ω ωw
Then, given a deformation amplitude a, the translational increment in the rigid block’s position ∆~x
and the angular increment in its orientation ∆φ can be computed as
∆~x = a~v
~ /||ω
~n = ω ~ ||2 , (1.12)
∆φ = a||ω~ ||2
where the rigid block’s rotation is described with a unit axis ~n passing though its COM and an angle
φ . Finally, we rewrite the the increment in the rigid block’s position ∆~x as a sum of two orthogonal
vectors,
∆~x = ∆~x⊥ + ∆~x|| , (1.13)
where ∆~x⊥ is orthogonal to ~n and ∆~x|| is collinear to ~n, and represent the ∆~x⊥ related motion as a
pure rotation about a new center~r0 , such that the final rigid blocks’s position is
~A′ = R(∆φ ,~n)(~A −~r0 ) +~r0 + ∆~x|| , (1.14)
where R(∆φ ,~n) is the rotation matrix describing rigid block’s rotation about an axis ~n by an angle
∆φ . See more details in [6].
6 Chapter 1. Theory
1.5 Potential function

In principle, our approach can be used with any potential function. Currently, to omit the need of
initial system’s energy minimization, we have only implemented an all-atom Anisotropic Network
Model (ANM) [7, 8], which is a type of elastic network model where the initial structure is always
at equilibrium. Figure 1.2 shows a schematic representation of this model. The all-atom ANM has
the following potential function,
γ
V (q) = ∑ (di j − di0j )2 , (1.15)
di0j <Rc
2
where di j is the distance between the ith and the jth atoms, di0j is the reference distance between
these atoms, as found in the original structure, γ is the stiffness constant, and Rc is a cutoff distance,
typically between 5 Å and 15 Å. The stiffness matrix corresponding to this potential function is
composed of the following blocks [2, 7, 8],
 0 2 0 0
(xi j ) xi j yi j xi0j z0i j

γ
Hi j = − 0 2  y0i j xi0j (y0i j )2 y0i j z0i j  i 6= j
(di j )
z0i j xi0j z0i j y0i j (z0i j )2 , (1.16)
Hii = − ∑ Hi j
j6=i
where xi j = xi − x j , yi j = yi − y j , and zi j = zi − z j . To rapidly compute this matrix, we use an

efficient neighbor search algorithm [9].
j
d~ij
Figure 1.2: Schematic representation of the elastic network model.

2. Program options
2.1 Usage
Typing the ’--help’ or ’-h’ flag produces the brief and more detailed description of the program
options,
NOLB --help
*******************************************************************
*-----------------------------------------------------------------*
*-----------NOLB : a Non-Linear Rigid Block NMA method------------*
*----------Authors: Alexandre Hoffmann & Sergei Grudinin----------*
*-Copyright (c): Nano-D team, Inria/CNRS Grenoble, France, 2018.--*
*--------------- e-mail: sergei.grudinin@inria.fr ----------------*
*---- http://team.inria.fr/nano-d/software/nolb-normal-modes/ ----*
*-----------------------------------------------------------------*
*******************************************************************
USAGE:
NOLB <pdb filename> <pdb reference file> [-o <output name>] [-n <number
of modes>] [--mode <unsigned integer>] ... [-c <cutoff distance>]
[-s <number of output frames>] [-a <maximum amplitude>] [-p
<potential function>] [-m] [--linear] [--trajectory] [--hetatm]
[--zdock <Zdock transform filename>] [--hex <Hex transform
filename>] [--frames <max number of input trajectory frames>]
[--covScaling <scaling of covalent interactions>] [-dcd <dcd file
name>] [--clust] [--coords <input traj file>] [--weights <input
weights file>] [-r <maximum rmsd>] [--dist <RMSD distribution>]
[--rand] [--nSteps <number of minimization steps>] [-t
<minimization tolerance>] [--blocks <Rigid blocks ids filename>]
[--analyze] [--collectivity] [--nlin] [--nIter <number of
additional Hessian matrix computations>] [-h] [--version] [--log]
Where:
<pdb filename>
(required) Input PDB file, can be a PDB multi-model trajectory file
<pdb reference file>

Reference PDB file. Only to compute the least RMSD motion. Disabled by
default.
-o <output name>, --output <output name>

Output file name. If nothing provided, a default name will be chosen.
8 Chapter 2. Program options
-n <number of modes>, --modes <number of modes>

Number of modes, 10 by default
-c <cutoff distance>, --cutoff <cutoff distance>

Interactions cutoff distance in angstroms, 5 by default
-s <number of output frames>, --solutions <number of output frames>

Number of frames in the output NMA trajectory, 10 by default. Please
note that s=0 suppresses the output.
-a <maximum amplitude>, --amplitude <maximum amplitude>

Maximum amplitude, 1.0 by default. The absolute amplitude value is
chosen automatically according to the system size. This option only
adjusts the absolute value.
-p <potential function>, --potential <potential function>

Potential function, 0 = Elastic Network, 1 = Gaussian Network.
Currently disabled.
-m, --minimize
Minimize obtained structures, off by default.
--linear
Additionally compute the linear modes. Off by default.
--trajectory
Save the output trajectory at equidistant points. Off by default.
--hetatm
Read heteroatoms from the input file. False by default.
--zdock <Zdock transform filename>

Zdock transforms file name. Experimental option.
--hex <Hex transform filename>

Hex transforms file name. Experimental option.
--frames <max number of input trajectory frames>

Max number of trajectory frames to use for trajectory analysis, 100 by
default.
--covScaling <scaling of covalent interactions>

Scaling of covalent interactions, 1 by default.
-dcd <dcd file name>, --dcd <dcd file name>

DCD file name. Requires corresponding PDB file.
--clust
Cluster trajectory frames, off by default.
--coords <input traj file>

Name of the text trajectory coordinate file that should correspond to
input pdb. Experimental option.
--weights <input weights file>

Name of the trajectory weights text file, where weights for different
frames are provided. Experimental option.
-r <maximum rmsd>, --rmsd <maximum rmsd>

Maximum rmsd for decoy generation. In case of MD clustering - RMSD
threshold.
--dist <RMSD distribution>

Power dependence of RMSD distribution of generated decoys on the input
RMSD, 0 for constant RMSD value (default), 1 for linear dependence,
0.5 for quadratic dependence, etc.
--rand
Random seed for decoy generation. Off by default.
2.1 Usage 9
--nSteps <number of minimization steps>

Number of minimization steps, 500 by default.
-t <minimization tolerance>, --tol <minimization tolerance>

Minimization tolerance, 0.1 by default
--blocks <Rigid blocks ids filename>

Rigid blocks ids file name.
--analyze
Analyze the computed modes. Off by default.
--collectivity
Prints modes collectivity. Off by default.
--nlin
Deterministic nonlinear transitions if the reference structure is
given, off by default.
--nIter <number of additional Hessian matrix computations>

Number of additional computations of the Hessian matrix if the
reference structure is given, 0 by default. If enabled, the
recommended value is 5.
-h, --help
Displays usage information and exits.
--version
Displays version information and exits.
--log
Displays ChangeLog information and exits.
Typing ’--version’ prints the current version of the program,

NOLB --version
NOLB version: 1.0, April 2018
Typing ’--log’ prints the changelog of the program,

NOLB --log
NOLB ChangeLog:
Version 0.1 from Feb 2016:
Initial release.
Version 0.2 from March 2016:
Added initial support of PDB trajectories.
Version 0.3 from Feb 2017:
Added reading of hetero atoms. Adapted for SAMSON GUI. Multiple tests performed.
Compiled on Linux, Win32 and MacOS.
Version 0.4 from May 2017:
Added automatic tests for convergence. More of noninear PCA support.
Version 0.5 from June 2017:
Added output of linear modes.
Version 0.6 from July 2017:
Added generation of NMA decoys.
Version 0.7 from December 2017:
Added possibility to manually specify rigid blocks.
Version 0.8 from February 2018:
Added support for DCD trajectories. Initial support of MD rapid clustering. Fixed support
of insertion codes when aligning the sequences.
Version 0.9 from March 2018:
Added basic analysis of the motions.
Version 1.0 from April 2018:
Nonlinear transitions added with the corresponding RMSD computations.

Version 1.1 from December 2018:
Randomization flag added.
2.2 Main options
<pdb filename> (required)

There is only one required argument, the path to the input PDB file. This can alternatively be a
PDB multi-model trajectory file. In this case, the NMA will be performed based on the covariance
matrix of the trajectory. A clustering of the trajectory can be additionally performed using the
corresponding flag.
<pdb reference file>

This options provides the path to the reference PDB file. It is only used to compute the least-RMSD
transition path. Disabled by default. The molecule may be different from the initial molecule. To
make the calculations robust, first a sequence alignment is performed.
-o <output name>, --output <output name>

This options provides the output file name. If nothing provided, a default name based on the name
of the input file will be chosen.
-n <number of modes>, --modes <number of modes>

This options provides the number of non-trivial modes to compute, 10 by default. If this number
exceeds the size of the Hessian matrix, it will be adapted accordingly.
-c <cutoff distance>, --cutoff <cutoff distance>

This options specifies the interaction cutoff distance for the elastic network models (in angstroms), 5
by default. The Hessian matrix is constructed according to this interaction distance. Some artifacts
should be expected for too short distances (< 5 Å).
-s <number of output frames>, --solutions <number of output frames>

This options provides the number of frames in the output NMA trajectory, 10 by default. The output
trajectory is written in the PDB file with multiple models. This flag specifies the number of models
in the output file. It also controls the number of models if pseudo-random decoys are generated.
-a <maximum amplitude>, --amplitude <maximum amplitude>

This options sets the maximum relative amplitude of NMA motion, 1.0 by default. The absolute
amplitude value is chosen automatically according to the system size. This flag only adjusts the
absolute value.
--linear
Flag to additionally compute the linear modes. Off by default. If computed, these are stored
separately.
2.2 Main options 11
--hetatm
Flag to read heteroatoms from the input file. False by default. Should be used for lipids, small
molecules, etc.
--trajectory
Flag to save the output trajectory at equidistant points. Off by default. If this option is disabled,
then the trajectory is written according to the natural harmonic motion of the oscillator.
--frames <max number of input trajectory frames>

This is the option to choose the maximum number of trajectory frames to use for trajectory analysis
of docking results, 100 by default. Only takes effect when analyzing the output of docking programs.
-r <maximum rmsd>, --rmsd <maximum rmsd>

If pseudo-random decoys are needed, this option specifies the maximum RMSD of the generated
decoys. If clustering of MD trajectories is required, this option will specify the clustering RMSD
threshold.
--dist <RMSD distribution>

This option defines the distribution of the decoys RMSD, 0 for constant RMSD (default), 0.5 for
quadratic dependence, 1 for linear dependence, etc.
--rand
Flag to reinitialize the random seed for decoy generation, off by default.
-m, --minimize
Flag to minimize the obtained structures, off by default.
--nSteps <number of minimization steps>

This option sets the number of minimization steps, 500 by default.
-t <minimization tolerance>, --tol <minimization tolerance>

This option sets the minimization tolerance, 0.1 by default.
--covScaling <scaling of covalent interactions>

This flag specifies the parameter by which the covalent interactions are scaled during the compu-
tation of the Hessian matrix. It does not significantly change the results and set to 1.0 by default.
-dcd <dcd file name>, --dcd <dcd file name>

Name of the trajectory file in the DCD format. If it is specified, the first PDB argument must
correspond to the topology of the DCD file.
--clust
This flag controls the clustering of the input trajectories. The user also needs to define the
corresponding RMSD clustering threshold. Off by default.
--analyze
This flag turns on the analysis of the computed normal modes. Currently, only the collectivities are
computed. Off by default.
--collectivity
This flag turns on outputting the collectivities of the modes. Off by default.
--nlin
This flag turns on a deterministic nonlinear transition computation if the reference structure is given.
No additional diagonalizations are required. Off by default.
--nIter
This flag specifies the number of additional computations of the Hessian matrix if the reference
structure is given, 0 by default. If enabled, the recommended value is 5. Will only take effect if the
--nlin flag is on.
--blocks <Rigid blocks ids filename>

Name of the text file where rigid blocks are specified. Currently, only residue IDs are supported.
Rigid blocks are separated by the "#" signs and each resID starts with a new line. All omitted
residues will be considered as individual rigid blocks. For example, the following file will add two
additional rigid blocks composed of residues with IDs 1-3 and 4-6:
#
1
2
3
#
4
5
6
-h, -–help
Flag to display usage information and exit.
--version
Flag to display version information and exit.
2.3 Experimental options 13
--log
Flag to display the ChangeLog information and exit.
2.3 Experimental options
–-zdock <Zdock transform filename>

This flag provides the path to the Zdock transforms file. Zdock protein docking poses are analyzed
then using the covariance matrix.
–-hex <Hex transform filename>

This flag provides the path to the Hex transforms file. Hex protein docking poses are analyzed then
using the covariance matrix.
–-coords <input traj file>

Name of the trajectory coordinate file that should correspond to the input pdb file. This one is used
for the analysis of the covariance matrix.
–-weights <input weights file>

Name of the trajectory weights text file, where weights for different frames are provided.
2.4 Currently suppressed options
-p <potential function>, –-potential <potential function>

Potential function, 0 = Elastic Network, 1 = Gaussian Network. Currently, the ANM model is only
used.
3. Examples
A 18 Å
B
0
D
18 Å
E F
5Å
5Å
Figure 3.1: Comparison of linear (A, C, E) and non-linear (B, D, F) normal modes computed for a
coiled coil protein (pdb code 2ch7). Three types of motions are shown, bending (A, B), stretching
(C, D), and twisting (E, F). Several snapshots at different deformation amplitudes are superposed to
each other. These are colored according to the values of the overall deformation, as measured by
the RMSD. The colorbars show the RMSD with respect to the initial position. The arrows follow
the trajectories of individual atoms.
3.1 Analysis of basic molecular motions 15
3.1 Analysis of basic molecular motions

We will start with the NOLB NMA of a mid-size coiled-coil protein from the cytoplasmic domain
of a bacterial chemoreceptor (pdb code 2ch7), which clearly demonstrates the three basic types of
molecular vibration motions. These are bending, stretching and twisting.
Example 1 To compute the first 10 non-trivial normal modes for a 2ch7.pdb protein structure,
type
NOLB 2ch7.pdb
This will produce the following output in the terminal (we suppose you are starting the program
from the terminal!),
*******************************************************************
*-----------------------------------------------------------------*
*-----------NOLB : a Non-Linear Rigid Block NMA method------------*
*----------Authors: Alexandre Hoffmann & Sergei Grudinin----------*
*-Copyright (c): Nano-D team, Inria/CNRS Grenoble, France, 2017.--*
*--------------- e-mail: sergei.grudinin@inria.fr ----------------*
*---- http://team.inria.fr/nano-d/software/nolb-normal-modes/ ----*
*-----------------------------------------------------------------*
*******************************************************************
=====================Reading input PDB file======================
Input PDB file................................................... : 2ch7.pdb
Number of chains read............................................ : 3
Number of atoms read............................................. : 4630
Found............................................................ : 4641 covalent bonds
................................................................. : 6264 angles
======================Constructing Hessian=======================
Number of disconnected atoms..................................... : 0
Number of interactions........................................... : 631042
Memory required for the Hessian matrix........................... : 129.991 Mb
============================Doing NMA============================
Memory required for the RTB projection matrix.................... : 1.27167 Mb
Reduced Hessian size : .......................................... : 3702 x 3702
Number of computed modes......................................... : 10
Null-space size.................................................. : 1
Lowest mode frequency............................................ : 0.000535409
Highest mode frequency........................................... : 0.00412211
Frequencies :
[ 1]................................................. : 0.000535409
[ 2]................................................. : 0.000542863
[ 3]................................................. : 0.00135327
[ 4]................................................. : 0.00153196
[ 5]................................................. : 0.00167521
[ 6]................................................. : 0.00261288
[ 7]................................................. : 0.00278308
[ 8]................................................. : 0.00347724
[ 9]................................................. : 0.003539
[10]................................................. : 0.00412211
====================Writing output PDB files=====================
Using output mask................................................ : 2ch7
Using output mask for nonlinear modes............................ : 2ch7_nlb_
=================================================================
=========================== Timing : ============================
=================================================================
Parsing arguments................................................ : 0.001036 s
Reading input PDB file........................................... : 0.015923 s
Constructing Hessian............................................. : 0.549772 s
Doing NMA........................................................ : 0.797914 s
Writing output PDB files......................................... : 0.775731 s
.................................................................
Total time : .................................................... : 2.14038 s
=================================================================
The first section lists the structural parameters, the number of atoms, chains, and the number of
16 Chapter 3. Examples
bonds and angles in the structure. These last two can be used during energy relaxation, which is
currently disabled. Water molecules are ignored and not listed, heteroatoms can be read with an
additional flag ’--hetatm’:
Input PDB file................................................... : 2ch7.pdb
................................................................. : 6264 angles
The second section lists parameters of the Hessian matrix. These are the number of interacting
atom pairs, which can be controlled with the ’--cutoff’ flag, the number of disconnected atoms,
which are stayed static, and the total memory required for the Hessian:
The next section lists parameters of the RTB basis and the diagonalization procedure. It provides
the size of the projection matrix, the size of the RTB Hessian, the size of the detected null-
space, the number of modes and the corresponding frequencies. These are the square roots of the
corresponding eigenvalues. The null-space is usually not detected fully, it should be 5 or 6 for
non-symmetric systems, and skipped from the further analysis:
============================Doing NMA============================
Null-space size.................................................. : 1
Frequencies :
[ 1]................................................. : 0.000535409
[ 2]................................................. : 0.000542863
[ 3]................................................. : 0.00135327
[ 4]................................................. : 0.00153196
[ 5]................................................. : 0.00167521
[ 6]................................................. : 0.00261288
[ 7]................................................. : 0.00278308
[ 8]................................................. : 0.00347724
[ 9]................................................. : 0.003539
[10]................................................. : 0.00412211
The next section lists the masks of the output files:

The output pdb trajectory files have the following names, ’2ch7_nlb_1.pdb’,’2ch7_nlb_2.pdb’,etc.
The number in the name refers to the normal mode id starting from 1. The output mask can be
changed with the ’-o newMask’ flag. Finally, the last section lists the timings of the program, split
into individual contributions:
=========================== Timing : ============================
=================================================================
Doing NMA........................................................ : 0.797914 s
3.2 Comparison with linear NMA 17
.................................................................
Total time : .................................................... : 2.14038 s
=================================================================
Here, the most time consuming operation is typically the diagonalization step. Writing trajectories
can be very long as well, depending on the number of required modes and frames in the trajectories.
Figure 3.1 shows some of the computed modes.
3.2 Comparison with linear NMA

Example 2 In this example we will compare the NOLB normal modes with the linear normal
modes. For this, we will enable the ’--linear’ flag, and will also output more trajectory frames
for a better visual perception of the motions using the ’-s 20’ flag. Finally, we will increase the
motion amplitude by a factor of 2 with the ’-a 2’ flag:
NOLB 2ch7.pdb -s 20 -a 2 --linear
The output will be only different when listing the output files,
Using output mask for linear modes............................... : 2ch7_linear_
Now, we have also computed the linear NMA trajectories stored in ’2ch7_linear_1.pdb’, ’2ch7_linear_2.pdb’,
etc. Figure 3.1 shows the visual difference between the computed linear and non-linear modes.
3.3 Another example with a bigger system that contains heteroatoms

Example 3 In this example we compute the first 50 normal modes for a larger system, the
photosystem II complex (pdb code 5b5e) with about 50,000 of atoms,
NOLB 5b5e.pdb -n 50
This will produce the following output, which we will cut for brevity,
Input PDB file................................................... : 5b5e.pdb
................................................................. : 57363 angles
============================Doing NMA============================
Frequencies :
[ 1]................................................. : 0.00156711
[ 2]................................................. : 0.00175741
=========================== Timing : ============================
=================================================================
Doing NMA........................................................ : 29.754 s
.................................................................
Total time : .................................................... : 65.8006 s

=================================================================
Please note that the PDB output now takes a significant part of the total execution time. This system
contains about 10,000 of small molecules, mostly lipids. In order to include these into the analysis,
we repeat the computations with the ’--hetatm’ flag,
Example 4 This example is identical to the previous one, with all the heteroatoms (excluding
water) included into the NMA calculations,
NOLB 5b5e.pdb -n 50 --hetatm,
This will produce the following output,

Input PDB file................................................... : 5b5e.pdb
................................................................. : 71368 angles
============================Doing NMA============================
Frequencies :
[ 1]................................................. : 0.00156711
[ 2]................................................. : 0.00175741
Please note the changed Hessian size and the spectrum of the normal modes. The visual inspection
of the first modes demonstrates that these are still the same. However, their order might have
changed because of the changed spectrum.
3.4 Finding the best transition between two protein conformations

In this example we are going to find the best direction in the NMA subspace that minimizes the
RMSD between two molecular structures. We should note that the structures are first aligned to each
other based on their sequence. Thus, the method only currently works with proteins.
Example 5 In this example we will compute the best non-linear trajectory and the correspond-
ing RMSD in the reduced NOLB subspace between two protein conformations. For the goal
(bound) conformation we choose the structure of the FAB 28 complex (pdb code 2hmi). For the
start conformation we choose the unbound structure of the HIV1 reverse transcriptase (pdb code
1s6p). Then, we will compute the best transition using the lowest 4 modes as follows,
NOLB 1s6p.pdb 2hmi.pdb -n 4 -s 20
The NOLB method will first align the two conformations,

Input PDB file................................................... : 1s6p.pdb
................................................................. : 11211 angles
Input PDB file................................................... : 2hmi.pdb
3.5 Analysis of an MD trajectory or NMR ensemble 19

................................................................. : 22422 angles
=====================Aligning input PDB file=====================
Total number of gaps ............................................ : 480
Alignement....................................................... :
-------------------------------------PISPIETVPVKLKPGMDGPKVKQ
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxPISPIETVPVKLKPGMDGPKVKQ
***********************
WY----------------------------------------------------------
WYQLEDIQMTQTTSSLSASLGDRVTISCSASQDISSYLNWYQQKPEGTVKLLIYYTSSLH
**
Then, it outputs the trajectory from the starting conformation to the goal using the optimal combina-
tion of the first 4 modes. It also prints the reduction in the overall RMSD for each mode,
Using output mask................................................ : 1s6p
Using output mask for nonlinear modes............................ : 1s6p_nlb_
RMSD reduction : ................................................ :
1s6p.pdb 3.621983 3.169015 2.894955 2.235346 2.196610
Initial RMSD..................................................... : 3.62198
Final RMSD....................................................... : 2.19661
Finally, it outputs the timings. We can see that overall it only takes about two seconds,
=========================== Timing : ============================
=================================================================
Aligning input PDB file.......................................... : 0.031428 s
Superposeing input PDB file...................................... : 0.000262 s
Doing NMA........................................................ : 1.16697 s
.................................................................
Total time : .................................................... : 2.40055 s
=================================================================
The obtained result can be visualized with popular software packages, for example, as
pymol 1s6p_nlb.pdb 2hmi.pdb
3.5 Analysis of an MD trajectory or NMR ensemble

In the next example, we will compute the principal components of an MD or NMR trajectory. This
will be done by the diagonalization of the covariance matrix. We will use a solution NMR structure
of a DNA (pdb code 5izp), which contains 15 models.
Example 6 To compute the first 10 non-trivial normal modes for an ensemble NMR structure,
type
NOLB 5izp.pdb -s 20 --linear
This produces the following output, cut for brevity,

Input PDB file................................................... : 5izp.pdb
This is a trajectory file........................................ :
=======================Reading Trajectory========================
Number of frames read............................................ : 15
===================Aligning Trajectory Frames====================

===============Constructing the Covariance matrix================
Memory Required for the Covariance Matrix........................ : 13.4146 Mb
=========================== Timing : ============================
=================================================================
Reading Trajectory............................................... : 0.005014 s
Aligning Trajectory Frames....................................... : 0.000192 s
Constructing the Covariance matrix............................... : 0.00419 s
Doing NMA........................................................ : 0.000838 s
.................................................................
Total time : .................................................... : 0.126667 s
=================================================================
You may see that the trajectory analysis is very fast, the only noticeable time here is actually the
output of the computed motions.
Example 7 In the following example, we will analyze a MD trajectory of a coiled-coil domain
of a chemoreceptor saved in DCD format. The length of the simulations was about 1 us, and the
frames were saved every 1 ns. To do so, we type
NOLB traj.pdb --dcd traj.dcd

Input PDB file................................................... : traj.pdb
=====================Reading DCD Trajectory======================
Trajectory File Name............................................. : traj.dcd
FRAMES........................................................... : 1040
NATOM............................................................ : 3546
LNFREAT.......................................................... : 3546
===================Aligning Trajectory Frames====================
===============Constructing the Covariance matrix================
Memory Required for the Covariance Matrix........................ : 863.396 Mb
============================Doing NMA============================
Lowest mode frequency............................................ : 5.44193e-05
=================================================================
=========================== Timing : ============================
=================================================================
Reading DCD Trajectory........................................... : 0.149615 s
Doing NMA........................................................ : 0.086615 s
.................................................................
Total time : .................................................... : 2.5702 s
=================================================================
Again, the trajectory analysis is very fast, and the principal components of a 1us simulations have
been computed in just a few seconds.
3.6 Clustering of MD trajectory frames 21
3.6 Clustering of MD trajectory frames
In the next example, we will provide a fast clustering method based on the RapidRMSD library
[10]. First, we will project the MD trajectory on the principal components, and then we will
use a fast method to compute RMSDs between the flexible conformations, if these are obtained
with a few collective motions. We will use a lysozyme MD trajectory (1960 atoms) with 10,000
frames.
Example 8 To cluster the frames of MD trajectory with RMSD threshold of 0.5 Å, type the
following,
NOLB lyz.pdb -dcd lyz.dcd --clust --rmsd 0.5

Input PDB file................................................... : lyz.pdb
=====================Reading DCD Trajectory======================
Trajectory File Name............................................. : lyz.dcd
TITLE............................................................ : Created by DCD plugin
NATOM............................................................ : 1960
============================Doing NMA============================
Frequencies :
[ 1]................................................. : 0.00203219
[ 2]................................................. : 0.00281031
[ 3]................................................. : 0.00587768
[ 4]................................................. : 0.00721599
[ 5]................................................. : 0.00790467
[ 6]................................................. : 0.00920148
[ 7]................................................. : 0.0119896
[ 8]................................................. : 0.013423
[ 9]................................................. : 0.0148008
[10]................................................. : 0.0172908
=====================Cluster the trajectory======================
Fast : Created 492 cluster(s) out of 10000 poses.
Clusters......................................................... :
0 1 2 3 4 5 6 7 8 9 10
...
9982 9985
9986 9987 9988 9990 9991 9998
=================================================================
=========================== Timing : ============================
=================================================================
Reading DCD Trajectory........................................... : 0.792337 s
Doing NMA........................................................ : 0.269174 s
Cluster the trajectory........................................... : 0.560206 s
.................................................................
Total time : .................................................... : 10.2181 s
You may see once again that the trajectory analysis including the subsequent clustering of a MD
trajectory is very fast.
3.7 Analysis of protein docking poses

In our next example, we will compute principal components of protein-protein docking poses.
We have chosen a standard example from the protein docking benchmark (pdb code 1d6r, with
the receptor 1D6R_r_b.pdb and the ligand 1D6R_l_b.pdb). We have computed the first 1,000 of
docking poses with the popular zDock package, Hex can be used as well here.
Example 9 To visualize the principal components of the computed docking poses type
NOLB outName.pdb --zdock ./1D6R.zdock.out
This produces the following output,

====================Reading Zdock Transforms=====================
========================Reading PDB files========================
Reading receptor PDB file........................................ : ./1D6R_r_b.pdb
Number of receptor chains read................................... : 1
Number of receptor atoms read.................................... : 1629
Reading ligand PDB file.......................................... : ./1D6R_l_b.pdb
Number of ligand chains read..................................... : 1
Number of ligand atoms read...................................... : 427
===================Reading Docking Trajectory====================
============================Doing NMA============================
Lowest mode frequency............................................ : 2.39176e-05
Frequencies :
[ 1]................................................. : 2.39176e-05
[ 2]................................................. : 5.74957e-05
[ 3]................................................. : 8.65304e-05
[ 4]................................................. : 0.000198992
[ 5]................................................. : 0.00031764
Using only....................................................... : 5 modes
Using output mask................................................ : outName
Using output mask for nonlinear modes............................ : outName_nlb
We should mention that this is just an example how to use the NOLB method for a simple PCA
analysis, which probably cannot be used for practical applications.
3.8 Structural ensembles

Another application of the NOLB method is the generation of structural ensembles (also called
’decoys’) with either a fixed deformation RMSD, or a certain distribution of these. Figure 3.2 shows
a sample output of the method.
Figure 3.2: Illustration of 100 decoys of a coiled-coil system 2ch7 protein with RMSD of 10
Å from the starting structure.
3.8 Structural ensembles 23
Example 10 In the first example we will generate 100 decoys of a coiled-coil system at 10
Å from the starting structure.
NOLB 2ch7.pdb -s 100 --rmsd 10
This creates a ’2ch7_nlb_decoys.pdb’ pdb trajectory file with 100 models with the following
output,
===================Writing output Decoy files====================
Number of files.................................................. : 100
Output RMSD...................................................... : 10 A
Using outputfilename............................................. : 2ch7_nlb_decoys.pdb
=================================================================
=========================== Timing : ============================
=================================================================
Doing NMA........................................................ : 0.605818 s
Writing output Decoy files....................................... : 0.848708 s
.................................................................
Total time : .................................................... : 1.53608 s
=================================================================
Example 11 In the second example we will generate 100 decoys of a the same system with
RMSD distributed linearly from 0 to 10 Å compared to the starting structure. We will also
minimize the obtained decoys.
NOLB 2ch7.pdb -s 100 --rmsd 10 --dist 1 -m
This creates a ’2ch7_nlb_decoys.pdb’ pdb trajectory file with 100 models with the following
output,
===================Writing output Decoy files====================
Number of files.................................................. : 100
Output RMSD...................................................... : 10 A
Using outputfilename............................................. : 2ch7_nlb_decoys.pdb
Current energy................................................... : 0.272494
N minimization steps............................................. : 0
Current Gradient norm............................................ : 0.0210788
Displacement .................................................... : 0.00167775
RMSD after minimization.......................................... : 2.45655
---
Current energy................................................... : 6.27796
N minimization steps............................................. : 1
Current Gradient norm............................................ : 0.0738622
Displacement .................................................... : 0.0135281
RMSD after minimization.......................................... : 7.13781
=================================================================
=========================== Timing : ============================
=================================================================
Doing NMA........................................................ : 0.603946 s
Writing output Decoy files....................................... : 2.1154 s
.................................................................
Total time : .................................................... : 2.79943 s
=================================================================
3.9 Structural transitions

Yet another application of the NOLB method is the generation of linear and nonlinear structural
transitions between different states of macromolecules (typically proteins) [11]. Figure 3.3 shows
a summary of these computed for the proteins from the Protein Docking Benchmark v5 [12].
RMSD, Å RMSD, Å
Figure 3.3: Transitions between the unbound (u) and bound (b) states of proteins from the Protein
Docking Benchmark v5 [12]. The top x-axis shows Cα RMSD between the two states. The bottom
x-axis lists the corresponding PDB codes of the complexes. The left plot shows the receptors, and
the right plot shows the ligands, as labelled by the authors of the benchmark. Only structures with
u-b RMSD ≥ 2 Å are shown. The y-axis shows the relative u-b transition that can be predicted
using the optimal linear combination of some number of normal modes. Results for the range
between 1 and 20 are shown in different colors (see the colorbar at the right).
Example 12 In the current example we will produce a transition between the unbound (u) and
the bound (b) forms of the ligand from the 3l89 complex.
NOLB 3L89_l_u.pdb 3L89_l_b.pdb
This creates a ’3L89_l_u_nlb.pdb’ pdb trajectory file for the best transition between the states, and
also the following output
Input PDB file................................................... : 3L89_l_u.pdb
Input PDB file................................................... : 3L89_l_b.pdb
=====================Aligning input PDB file=====================
Total number of gaps ............................................ : 0
Alignement....................................................... :
CEEPPTFEAMELIGKPKPYYEIGERVDYKCKKGYFYIPPLATHTICDRNHTWLPVSDDAC
CEEPPTFEAMELIGKPKPYYEIGERVDYKCKKGYFYIPPLATHTICDRNHTWLPVSDDAC
************************************************************
YRETCPYIRDPLNGQAVPANGTYEFGYQMHFICNEGYYLIGEEILYCELKGSVAIWSGKP
YRETCPYIRDPLNGQAVPANGTYEFGYQMHFICNEGYYLIGEEILYCELKGSVAIWSGKP
************************************************************
3.9 Structural transitions 25
PICEKV
PICEKV
******
===================Superposeing input PDB file===================

Interaction cutoff distance...................................... : 5 A
All-atom Hessian size : ......................................... : 3054 x 3054
============================Doing NMA============================
Null-space size.................................................. : 1
Frequencies :
[ 1]................................................. : 0.000266116
[ 2]................................................. : 0.000420865
[ 3]................................................. : 0.000595849
[ 4]................................................. : 0.00120617
[ 5]................................................. : 0.00132804
[ 6]................................................. : 0.00146345
[ 7]................................................. : 0.00150491
[ 8]................................................. : 0.00159594
[ 9]................................................. : 0.00170802
[10]................................................. : 0.00198693
Using output mask................................................ : 3L89_l_u
Using output mask for nonlinear modes............................ : 3L89_l_u_nlb_
Number of aligned atoms.......................................... : 126
RMSD reduction : ................................................ :
4.445432 2.258740 2.177358 1.470716 1.412096 1.411807 1.337768 1.337624
1.325643 1.293560 1.182593
Initial RMSD..................................................... : 4.44543
Final RMSD....................................................... : 1.18259
Reduction in RMSD by............................................. : 0.733976
=================================================================
=========================== Timing : ============================
=================================================================
Aligning input PDB file.......................................... : 0.000866 s
Superposeing input PDB file...................................... : 4.9e-05 s
Doing NMA........................................................ : 0.05935 s
.................................................................
Total time : .................................................... : 0.096657 s
=================================================================
Please note, that here we have found a reduction of RMSD by 73% if the best combination of the
first ten modes is used. This analysis can be performed on the whole benchmark, for example, with
the following bash script,
for f in b5/*_r_u.pdb; do NOLB $f ${f:0:${#f}-5}b.pdb -n 20 >> results-R-20modes.txt; done
for f in b5/*_l_u.pdb; do NOLB $f ${f:0:${#f}-5}b.pdb -n 20 >> results-L-20modes.txt; done
And the results can be presented as a plot (similar to the one in Figure 3.3) using the following
python-based post-processing,
python benchmark.py results-R-20modes.txt
Where the ’benchmark.py’ script is

#benchmark.py to print the RMSD reduction results

import sys, os
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
import matplotlib.cm as cm
import collections
inputFile = sys.argv[-1]
with open(inputFile) as f:
lines = f.readlines()
inputfiles = []
rmsds = []
frequencies = []
globalData = {}
startRMSD = 0
for line in lines:
words = line.split()
if (len(words)==0) : continue
if (words[0] == "*******************************************************************") :
if (len(rmsds)) :
globalData[rmsds[0]] = [inputfiles, rmsds]
inputfiles = []
frequencies = []
rmsds = []
if (words[0]=="Input"):
inputfiles.append(words[4].split(’/’)[-1])
if (words[0][0]=="["):
frequencies.append(float(words[-1]))
if (startRMSD) :
startRMSD = 0
for w in words:
rmsds.append(float(w))
if (words[0]=="RMSD"):
startRMSD = 1
globalData = collections.OrderedDict(sorted(globalData.items()))
cutoff = 2.0
nModes=20
Quality = np.zeros(nModes)
QualityAbs = np.zeros(nModes)
lenQ = 0
columns = []
labels = []
rows = []
expectedLength = len(list(globalData.values())[0][1])
print("expectedlength = ", expectedLength)
for dat in globalData:
rmsds = globalData[dat][1]
initRMSD = rmsds[0]
if (initRMSD < cutoff or initRMSD > 5) : continue
lenQ += 1
for idx, r in enumerate(rmsds[1:nModes+1]):
Quality[idx] += (initRMSD - r) / initRMSD
QualityAbs[idx] += (initRMSD - r)
if (initRMSD < cutoff) : continue

columns.append(globalData[dat][0][0][:4].lower()) # column name
labels.append("%.1f"%(initRMSD))
previousRmsd = initRMSD
3.10 Nonlinear structural transitions 27
reductions = []
while(len(rmsds) < expectedLength) : rmsds.append(rmsds[-1])
for idx, r in enumerate(rmsds[1:nModes+1]):
reduction = previousRmsd - r
previousRmsd = r
relativeReduction = reduction / initRMSD
if (len(rows)<idx+1) : rows.append([])
rows[idx].append(relativeReduction)
index = np.arange(len(columns)) + 0.3

bar_width = 1.0
print ("Quality, Abs Quality")

for idx,q in enumerate(Quality):
print (idx+1, q/lenQ, QualityAbs[idx] / lenQ)
exit
print (columns)
print (labels)
lenRows = len(rows)
colormap = plt.cm.Vega20b
colors = colormap(np.linspace(0, 1, lenRows))
normalize = mcolors.Normalize(vmin=0, vmax=lenRows)
# Initialize the vertical-offset for the stacked bar chart.

y_offset = np.array([0.0] * len(columns))
# Plot bars and create text labels for the table

fig = plt.figure()
ax1 = fig.add_subplot(111)
ax2 = ax1.twiny()
cell_text = []
for idx, row in enumerate(rows):
plt.bar(index, row, bar_width, bottom=y_offset, color=colors[idx])
y_offset = y_offset + row
cell_text.append([’%1.1f’ % (x/1000.0) for x in y_offset])
# setup the colorbar
ax1.set_ylabel("RMSD reduction", size=12)

ax1.set_xlabel("Ligand code", size=12)
ax2.set_xlabel("RMSD, $\t{\AA}$", size=12)
ax1.set_xticks(index)
ax2.set_xticks(index)
ax1.set_xticklabels(columns, rotation=90, size=10)

ax2.set_xticklabels(labels, rotation=90, size=10)
ax2.set_xlim(plt.axis()[:2])
print (plt.axis())
print (ax1.set_xlim(plt.axis()[0],plt.axis()[1]*1.26))
ax1.set_frame_on(False)
scalarmappaple = plt.cm.ScalarMappable(norm=normalize, cmap=colormap)

scalarmappaple.set_array(index)
plt.colorbar(scalarmappaple,ticks=range(1,nModes+1))
plt.show()
3.10 Nonlinear structural transitions

Here we demonstrate the way to produce deterministic nonlinear transitions as in the previous
example.
Example 13 In this example we produce a nonlinear transition between the unbound (u) and
the bound (b) forms of the ligand from the 3l89 complex.
NOLB 3L89_l_u.pdb 3L89_l_b.pdb -n 7 --nlin
This creates a ’3L89_l_u_nlb.pdb’ pdb trajectory file for the best nonlinear transition between the
states, and also the following output (cut for brevity) :
=========================RMSD statistics=========================
RMSD reduction : ................................................ :
4.445432 2.258740 2.177358 1.470716 1.412096 1.411807 1.337768 1.337624
Amplitudes :
454.891 -76.1758 197.929 -39.3717 -7.06865 83.5584 3.02367
RMSD reduction : ................................................ :
4.041845 2.112556 2.041760 1.442365 1.394135 1.393912 1.333585 1.333479
RMSD reduction : ................................................ :
3.680333 1.982142 1.920188 1.412448 1.372653 1.372480 1.323201 1.323135
...
...
RMSD reduction : ................................................ :
1.192290 1.189762 1.189452 1.189097 1.189088 1.189087 1.188989 1.188918
RMSD reduction : ................................................ :
1.192304 1.190237 1.189976 1.189690 1.189683 1.189683 1.189598 1.189536
Initial RMSD..................................................... : 4.44543
Final RMSD....................................................... : 1.18954
Reduction in RMSD by............................................. : 0.732414
The RMSD reduction values in lines correspond to the analytical estimation of these for the linear
case. The first value in the column correspond to the nonlinear case. We see that the final nonlinear
RMSD to the target is 1.19 Å, whereas the final linear RMSD to the target is 1.34 Å.
3.11 Analysis of modes

To analyse the obtained normal modes, one can use different criteria, for example, the reduction of
the RMSD between two states of a molecular system. Another characteristic is the collectivity κ ,
which is proportional to the exponential of the “information entropy” of the eigenvector qi [13]:
( )
1 N
2 2
κi = exp − ∑ qi,n log qi,n , (3.1)
N i=1
where the sum runs over all the N atoms in the molecule, and
q2i,3n−2 + q2i,3n−1 + q2i,3n

q2i,n = α , (3.2)
mn
with the normalization factor α taken such that ∑i q2i,n = 1. N κ gives an effective number of
nonzero eigenvector components q2i,n . Thus, κ is confined to the interval {N −1 ;1}. If κ1 = 1, then
the corresponding mode is maximally collective and has all the identical amplitudes q2i,n , which
happens for rigid-body motion, for example. In the limit of extreme local motion, where a mode
affects a single atom only, κ1 = 1 is minimal and equals to 1/N .
Example 14 In the following example we will compute collectivities for the first twenty modes
of the coiled-cil protein 2ch7.
NOLB 2ch7.pdb --analyze -n 20 -s 0
This creates the following output (cut for brevity),

3.11 Analysis of modes 29
============================Doing NMA============================
=================================================================
NMA collectivities :
[ 1].......................................... : 0.682681
[ 2].......................................... : 0.682069
[ 3].......................................... : 0.638731
[ 4].......................................... : 0.670347
[ 5].......................................... : 0.723305
[ 6].......................................... : 0.740871
[ 7].......................................... : 0.746425
[ 8].......................................... : 0.647954
[ 9].......................................... : 0.741014
[10].......................................... : 0.764615
[11].......................................... : 0.760854
[12].......................................... : 0.611267
[13].......................................... : 0.754941
[14].......................................... : 0.718023
[15].......................................... : 0.594634
[16].......................................... : 0.711807
[17].......................................... : 0.5525
[18].......................................... : 0.760569
[19].......................................... : 0.781436
[20].......................................... : 0.716035
=================================================================
We can see that the modes 10, 11, 13, 18, 19 are the most collective, and the modes 15 and 17 are
the least collective.
4. Related methods
4.1 GUI
A graphical user interfaces created for the SAMSON software platform is available at https:
//www.samson-connect.net. Figure 4.1 shows the current version of this GUI.
Figure 4.1: NOLB graphical user interface for the SAMSON software platform.
4.2 Morphing paths

The NOLB normal modes can be used to produce morphing pathways between multiple conforma-
tions of biological molecules. It is not a well tested approach yet. However, in the future it can be
merged with the NOLB program.
4.3 SAXS-guided structure optimization 31
4.3 SAXS-guided structure optimization

We have been also using the NOLB normal modes as a low-dimensional representation of the
protein motion subspace in the recent CASP12 blind structure prediction exercise. More precisely,
we used for in the SAXS-assisted protein targets, where we numerically computed the gradient of
the SAXS loss function and optimized the structures towards the gradients in this low-dimensional
space. This method will be distributed separately as a part of our Pepsi-SAXS scattering package at
https://team.inria.fr/nano-d/software/pepsi-saxs/.
Bibliography
[1] E Bright Wilson, J C Decius, and Paul C Cross. Molecular Vibrations: The Theory of Infrared
and Raman Spectra. McGraw-Hill, 1955 (cited on page 3).
[2] Ivet Bahar et al. “Normal Mode Analysis of Biomolecular Structures: Functional Mechanisms
of Membrane Proteins”. In: Chem. Rev. 110.3 (Mar. 2010), pages 1463–1497 (cited on
pages 3, 6).
[3] Philippe Durand, Georges Trinquier, and Yves-Henri Sanejouand. “A New Approach for
Determining Low-Frequency Normal Modes in Macromolecules”. In: Biopolymers 34.6
(June 1994), pages 759–771 (cited on page 3).
[4] Florence Tama et al. “Building-Block Approach for Determining Low-Frequency Normal
Modes of Macromolecules”. In: Proteins: Struct., Funct., Bioinf. 41.1 (2000), pages 1–7
(cited on page 3).
[5] Timothy R Lezon et al. “Elastic Network Models for Biomolecular Dynamics: Theory and
Application to Membrane Proteins and Viruses”. In: Handbook on Biological Networks.
World Scientific Pub Co Pte Lt, Dec. 2009, pages 129–158 (cited on page 4).
[6] Alexandre Hoffmann and Sergei Grudinin. “NOLB: Nonlinear Rigid Block Normal-Mode
Analysis Method”. In: J. Chem. Theory Comput. 13.5 (2017), pages 2123–2134 (cited on
page 5).
[7] Pemra Doruker, Ali Rana Atilgan, and Ivet Bahar. “Dynamics of Proteins Predicted by
Molecular Dynamics Simulations and Analytical Approaches: Application to α -Amylase
Inhibitor”. In: Proteins: Struct., Funct., Bioinf. 40.3 (2000), pages 512–524 (cited on page 6).
[8] A R Atilgan et al. “Anisotropy of Fluctuation Dynamics of Proteins with an Elastic Network
Model”. In: Biophys. J. 80.1 (Jan. 2001), pages 505–515 (cited on page 6).
[9] Svetlana Artemova, Sergei Grudinin, and Stephane Redon. “A Comparison of Neigh-
bor Search Algorithms for Large Rigid Molecules”. In: J. Comput. Chem. 32.13 (2011),
pages 2865–2877 (cited on page 6).
[10] Emilie Neveu et al. “RapidRMSD : Rapid determination of RMSDs corresponding to motions
of flexible molecules”. In: Bioinformatics (2018) (cited on page 21).
[11] Sergei Grudinin and Alexandre Hoffmann. “Critical assessment of macromolecular confor-
mational transitions computed via linear and nonlinear normal mode analysis”. Unpublished
(cited on page 24).
BIBLIOGRAPHY 33
[12] Thom Vreven et al. “Updates to the Integrated Protein-Protein Interaction Benchmarks:
Docking Benchmark Version 5 and Affinity Benchmark Version 2.” In: J. Mol. Biol. 427.19
(Sept. 2015), pages 3031–3041 (cited on page 24).
[13] Rafael Brüschweiler. “Collective protein dynamics and nuclear spin relaxation”. In: The
Journal of chemical physics 102.8 (1995), pages 3396–3403 (cited on page 28).
Index
A I
Analysis of basic molecular motions . . . . . . 15 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

Analysis of modes . . . . . . . . . . . . . . . . . . . . . . 28
Analysis of protein docking poses . . . . . . . . 22 M
Analysis of trajectories . . . . . . . . . . . . . . . . . . 19
Main options . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Morphing paths . . . . . . . . . . . . . . . . . . . . . . . . . 30
C
N
Clustering of MD trajectory frames . . . . . . . 21
Collectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 NMA theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Comparison with linear NMA . . . . . . . . . . . . 17 nonlinear structural transitions . . . . . . . . . . . . 27
Currently suppressed options . . . . . . . . . . . . . 13
P
D
Protein Docking Benchmark . . . . . . . . . . . . . 24
Decoys. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .22
R
E RTB projection method . . . . . . . . . . . . . . . . . . . 4
Example with a bigger system that contains
heteroatoms . . . . . . . . . . . . . . . . . . . . 17 S
Experimental options . . . . . . . . . . . . . . . . . . . . 13
SAXS-guided structure optimization . . . . . . 31
Structural ensembles . . . . . . . . . . . . . . . . . . . . 22
F Structural transitions . . . . . . . . . . . . . . . . . . . . 24
Finding the best transition between two protein T
conformations . . . . . . . . . . . . . . . . . . 18
The NOLB method . . . . . . . . . . . . . . . . . . . . . . . 5
G
U
GUI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
H
Heteroatoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

Nolb Manual

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Nolb Manual

Încărcat de

Drepturi de autor:

Formate disponibile

User Guide

NOLB : Non-Linear Normal Mode Analysis, version 1.1, December 2018

c 2018 Sergei Grudinin

1.2 NMA theory

M(q̈ + q¨0 ) + ∇V (q0 + q) ≈ M q̈ + Kq = 0, (1.1)

1.3 RTB projection method

P = ... ... ... . (1.7)

1.4 The NOLB method

1.5 Potential function

where xi j = xi − x j , yi j = yi − y j , and zi j = zi − z j . To rapidly compute this matrix, we use an

Figure 1.2: Schematic representation of the elastic network model.

<pdb reference file>

-o <output name>, --output <output name>

-n <number of modes>, --modes <number of modes>

-c <cutoff distance>, --cutoff <cutoff distance>

-s <number of output frames>, --solutions <number of output frames>

-a <maximum amplitude>, --amplitude <maximum amplitude>

-p <potential function>, --potential <potential function>

--zdock <Zdock transform filename>

--hex <Hex transform filename>

--frames <max number of input trajectory frames>

--covScaling <scaling of covalent interactions>

-dcd <dcd file name>, --dcd <dcd file name>

--coords <input traj file>

--weights <input weights file>

-r <maximum rmsd>, --rmsd <maximum rmsd>

--dist <RMSD distribution>

--nSteps <number of minimization steps>

-t <minimization tolerance>, --tol <minimization tolerance>

--blocks <Rigid blocks ids filename>

--nIter <number of additional Hessian matrix computations>

Typing ’--version’ prints the current version of the program,

NOLB version: 1.0, April 2018

Typing ’--log’ prints the changelog of the program,

Nonlinear transitions added with the corresponding RMSD computations.

2.2 Main options

<pdb filename> (required)

<pdb reference file>

-o <output name>, --output <output name>

-n <number of modes>, --modes <number of modes>

-c <cutoff distance>, --cutoff <cutoff distance>

-s <number of output frames>, --solutions <number of output frames>

-a <maximum amplitude>, --amplitude <maximum amplitude>

--frames <max number of input trajectory frames>

-r <maximum rmsd>, --rmsd <maximum rmsd>

--dist <RMSD distribution>

--nSteps <number of minimization steps>

-t <minimization tolerance>, --tol <minimization tolerance>

--covScaling <scaling of covalent interactions>

-dcd <dcd file name>, --dcd <dcd file name>

--blocks <Rigid blocks ids filename>

2.3 Experimental options

–-zdock <Zdock transform filename>

–-hex <Hex transform filename>

–-coords <input traj file>

–-weights <input weights file>

2.4 Currently suppressed options

-p <potential function>, –-potential <potential function>

3.1 Analysis of basic molecular motions

The next section lists the masks of the output files:

3.2 Comparison with linear NMA

3.3 Another example with a bigger system that contains heteroatoms

Total time : .................................................... : 65.8006 s

This will produce the following output,